Аңдатпа
Kazakhstan has prioritised digitalising its criminal justice processes, implementing AI-driven tools such as automated forensic systems and "digital autopsy" platforms to increase efficiency and transparency. However, challenges persist: interrogation recordings can still be falsified or incomplete due to inadequate technical equipment and procedural gaps, undermining evidence integrity. This study investigates a combined AI-blockchain framework to enhance the reliability of verbal investigative actions. It examines an integrated system in which automatic speech recognition (ASR) transcribes interrogation audio, and a private blockchain ledger immutably timestamps each transcript. The study employs a mixed-methods approach: data include recorded criminal interrogation sessions (audio transcripts) and relevant legal documents. A neural ASR model was trained on these transcripts (approximately 50 sessions) to produce digital text, and standard analytical methods were applied to evaluate system performance. Simultaneously, each transcript was recorded on a blockchain registry, and metrics such as transcription accuracy and error rate were measured. Results show the proposed system attains high transcription accuracy (94% word recognition) and significantly reduces manual review time (by about 40%). In experiments, word error rates dropped from roughly 10% (baseline) to 5% with the ASR model, and the blockchain ledger recorded 100% of transcripts with negligible overhead. The findings suggest that integrating AI-driven transcription with blockchain anchoring markedly enhances the fidelity of investigative recordings. Quantitatively, the AI-blockchain platform halved transcription errors and enabled reliable reconstruction of interrogation timelines. These results indicate the combined approach substantially improves the reliability and integrity of verbal evidence, reinforcing the objective completeness of recorded testimony.